Why does research require so much computer power?

[Matias_Chambers]

I first figured out what Folding@home is right now. I have decided to try it as most of the time when I'm just web browsing, I'm only using 0-10% CPU usage and even at 100% CPU usage and 100% RAM usage I barely notice it unless I'm gaming. So I figured why not try it out. My question is, why does scientific research or whatever require so much computer power? 

[Oshino Shinobu]

If you read some of the science that is being done by F@H, you can get an idea of why it takes so much processing power. 

 

http://folding.stanford.edu/home/the-science/

 

Essentially, simulating folding proteins takes a very long time as it's extremely complex.

 

"Folding is a very complex process, and it’s often challenging to study in the laboratory. It’s amazing that not only do proteins self-assemble — fold — but they do so amazingly quickly: some as fast as a millionth of a second. While this time is very fast on a person’s timescale, it’s remarkably long for computers to simulate. In fact, even modern computers can take a day to simulate about 50 nanoseconds (50/1,000,000,000 of a second). Unfortunately, many proteins fold on the millisecond timescale (1,000,000 nanoseconds). Thus, it would take 20,000 days to simulate folding — i.e. it would take 60 years! That’s a long time to wait for one result!"

 

https://en.wikipedia.org/wiki/Protein_folding

 

EDIT: With F@H, you're only performing part of the folding calculation. Doing all of it on one PC would take a very, very long time to simulate. 

[corrado33]

Turns out modeling atoms is hard! The thing is, we can model all of the folding systems in less accurate ways on any computer, easily. However, if you want SUPER accurate results that take into account ALL of the forces that could change how a protein folds, then you have to use more computationally expensive methods. 

 

Look at it this way. Using a very expensive computational method, I can run a simulation that takes MONTHS to finish on a 9 atom system running on a supercomputer using 8 cores and 16 GB of ram. 9 FREAKING ATOMS. Proteins are very massive comparatively. Very rough calculations show that the average protein in the human body contains 10,000 atoms.

 

Now, they certainly don't use the same methods, but if they could, you could bet your butt that the scientists would use those more accurate (and computationally expensive) methods. 

 

It's all about shells of... let's call it "accuracy." If you're modeling a small molecule, less than a dozen atoms or so, you can model each atom INDIVIDUALLY. Model the forces imposed on each OTHER atom by each atom. Model the electrons in the shells and the forces they generate. You can solve the math exactly (well exact enough within the limits of the uncertainty principle.) 

 

When you get to bigger molecules, you start to make assumptions. You assume that each atom generates such and such a force, without actually calculating that force. When you get to bigger and bigger systems, you start working with larger and larger assumptions simply because we don't have the computational power to use those better methods. Say, for example, you're working with a solvent. Most of the time you can model the solvent using a gradient of force. The solvent molecules further and further away will generate less and less force on the molecule you're modeling.

 

Basically the reason they need so much computing power is because they want more accurate answers.

 

Also note: Supercomputers aren't really any more powerful than your or my computer. They just happen to have TONS of CPUs and TONS of memory. When I did computational chemistry, we bought a bunch of mac pros that ran the jobs faster than our local supercomputing center. But, supercomputers allow the use of many (or all) of the resources (CPUs, Memory) to be used at once for things like weather simulations that are built to take advantage of it. The largest jobs require the most resources, so there were simply jobs that we could not run on our mac pros due to resource limitations.

[Ezzy-525]

1 minute ago, wrathoftheturkey said:

We're trying to examine how hundreds and thousands of atoms come together and fold into the three dimensional structures proteins need to do what they do. Nvidia PhysX ain't got nothing on this level of physics.

OMG AMD fanboy alert!

 

But seriously, as others have said, distributed computing is needed as the sheer computational power needed is beyond any one machine. In essence it's like one scientist trying to do a project completely alone...more scientists helps speed things up.

[corrado33]

Not to mention there are most likely millions (or more) of proteins that need to be studied.

[Ithanul]

Yep, there is a reason distributed computing and supercomputers have their use.

Simulations of extreme large interactions or extremely small interactions require a large amount of number crunching and variables.  Especially if you want to get the most precise calculations and measurements.